AWS RDS Instance Recommendation 10 million views

General questions
1

Hello,

We are looking at setting up our stats server on AWS. We are aiming to handle 10 Million Pageviews. I wanted to get some input on which RDS instance everyone uses. For example, If I should go for a t3 or an m5? For EC2 I’m going for an m5.2xlarge.

We currently are running a Linode server (12 CPU/32gb ram) with matomo and mysql all in one.

Any input appreciated.

Thanks

2

Hi James,

10M pageview per what timeframe? If you do that per year, you should be fine with a single node… haven’t done my math to be honest. If you do 10M per day, then you need a different setup than you now use.

I have just finished a major redo of our high traffic matomo instance. Now running at 8-10M per day easy and peaked at 36M a couple of weeks ago. I do have some tips, but then you need to tell me a little more.

You can also read: How to configure Matomo for speed User Guide - Analytics Platform - Matomo

We have ended up running with REDIS and a MasterSlave configuration. And optimised the configuration by splitting up the data capture from the data processing node.

Need more information, just let me know, I might just be able to share some stuff that helped me.
Regards,
Robert

3

Thanks for the reply. Sorry about that. 10 million page views per month.

4

A 10M per month should be on a xlarge db and ws instance. As long as you do not have large traffic spikes you should be fine.

5

Hi Robert, are you using any particular RDS parameters you might recommend? I have a db.r5.large writer and two db.t3.mediums readers for ~ 2M visits/month which are having a hard time on the archiving side for around 15 segments and 10 goals. Lots of “server gone away” in the logs and general slowness in the UI.

6

Hi Daniel

We are not using AWS. But an on prem solution. But our reader database is scaled to 32 cores and 128GB memory. With that we have scaled the segments to over 150 segments. But we do struggle with this amount too. The more segments we use, more database power has been needed.

Also some segments are just problematic queries. So you have to take a close look at the segments too.

Hope this helps
Robert

7

Thank you Robert. It does help to have a sense of the scaling. Any viable alternative to segments you recommend? Thanks!